Abstract: A new method based on GMM+SVM algorithm is proposed to determine the best dividing line for crack classification in acoustic emission parameter analysis. Use Gaussian mixture model (GMM) to cluster the characteristics of the acoustic emission signal obtained in the test process to obtain the label data, and then the SVM hyperplane dividing line is obtained by the cross clusters in the hyperplane separating data created by the support vector machine (SVM) algorithm, the distinction between tensile and shear zones in crack classification are optimized. Taking granite as the research object, the acoustic emission signals of granite specimens during fracture process are studied by uniaxial loading test. The analysis results show that the GMM+SVM hyperplane algorithm can accurately distinguish the cross clusters between the tension and shear regions. The granite is dominated by tension cracks in the early and mid stages of the entire stress loading process, and the proportion of shear cracks in the later period continues to increase. The proportion of shear cracks changes abruptly at 80% to 90% peak stress stage. It is finally verified that this method can correspond to the different failure modes of granite during the whole stress loading process, and can provide a basis for the prediction of rock instability and failure.